[feat] hypamemory first commit
This commit is contained in:
kwaroran
139
src/ts/process/memory/hypamemory.ts
Normal file
139
src/ts/process/memory/hypamemory.ts
Normal file
@@ -0,0 +1,139 @@
|
||||
import localforage from "localforage";
|
||||
import { similarity } from "ml-distance";
|
||||
import { globalFetch } from "src/ts/storage/globalApi";
|
||||
|
||||
|
||||
export class HypaProcesser{
|
||||
oaikey:string
|
||||
vectors:memoryVector[]
|
||||
forage:LocalForage
|
||||
|
||||
constructor(){
|
||||
this.forage = localforage.createInstance({
|
||||
name: "hypaVector"
|
||||
})
|
||||
}
|
||||
|
||||
async embedDocuments(texts: string[]): Promise<number[][]> {
|
||||
const subPrompts = chunkArray(texts,512);
|
||||
|
||||
const embeddings: number[][] = [];
|
||||
|
||||
for (let i = 0; i < subPrompts.length; i += 1) {
|
||||
const input = subPrompts[i];
|
||||
|
||||
const data = await this.getEmbeds(input)
|
||||
|
||||
embeddings.push(...data);
|
||||
}
|
||||
|
||||
return embeddings;
|
||||
}
|
||||
|
||||
|
||||
async getEmbeds(input:string[]|string) {
|
||||
const gf = await globalFetch("https://api.openai.com/v1/embeddings", {
|
||||
headers: {
|
||||
"Authorization": "Bearer " + this.oaikey
|
||||
},
|
||||
body: {
|
||||
"input": input,
|
||||
"model": "text-embedding-ada-002"
|
||||
}
|
||||
})
|
||||
const data = gf.data
|
||||
|
||||
|
||||
if(!gf.ok){
|
||||
throw gf.data
|
||||
}
|
||||
|
||||
const result:number[][] = []
|
||||
for(let i=0;i<data.data.length;i++){
|
||||
result.push(data.data[i].embedding)
|
||||
}
|
||||
|
||||
return result
|
||||
}
|
||||
|
||||
|
||||
|
||||
async addText(texts:string[]) {
|
||||
|
||||
for(let i=0;i<texts.length;i++){
|
||||
const itm:memoryVector = await this.forage.getItem(texts[i])
|
||||
if(itm){
|
||||
itm.alreadySaved = true
|
||||
this.vectors.push(itm)
|
||||
}
|
||||
}
|
||||
|
||||
texts = texts.filter((v) => {
|
||||
for(let i=0;i<vectors.length;i++){
|
||||
if(this.vectors[i].content === v){
|
||||
return false
|
||||
}
|
||||
}
|
||||
return true
|
||||
})
|
||||
|
||||
if(texts.length === 0){
|
||||
return
|
||||
}
|
||||
const vectors = await this.embedDocuments(texts)
|
||||
|
||||
const memoryVectors:memoryVector[] = vectors.map((embedding, idx) => ({
|
||||
content: texts[idx],
|
||||
embedding
|
||||
}));
|
||||
|
||||
for(let i=0;i<memoryVectors.length;i++){
|
||||
const vec = memoryVectors[i]
|
||||
if(!vec.alreadySaved){
|
||||
await this.forage.setItem(texts[i], vec)
|
||||
}
|
||||
}
|
||||
|
||||
this.vectors = memoryVectors.concat(this.vectors)
|
||||
}
|
||||
|
||||
async similaritySearch(query: string) {
|
||||
const results = await this.similaritySearchVectorWithScore((await this.getEmbeds(query))[0],);
|
||||
|
||||
return results.map((result) => result[0]);
|
||||
}
|
||||
|
||||
async similaritySearchVectorWithScore(
|
||||
query: number[],
|
||||
): Promise<[string, number][]> {
|
||||
const memoryVectors = this.vectors
|
||||
const searches = memoryVectors
|
||||
.map((vector, index) => ({
|
||||
similarity: similarity.cosine(query, vector.embedding),
|
||||
index,
|
||||
}))
|
||||
.sort((a, b) => (a.similarity > b.similarity ? -1 : 0))
|
||||
|
||||
const result: [string, number][] = searches.map((search) => [
|
||||
memoryVectors[search.index].content,
|
||||
search.similarity,
|
||||
]);
|
||||
|
||||
return result;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
type memoryVector = {
|
||||
embedding:number[]
|
||||
content:string,
|
||||
alreadySaved?:boolean
|
||||
}
|
||||
|
||||
const chunkArray = <T>(arr: T[], chunkSize: number) =>
|
||||
arr.reduce((chunks, elem, index) => {
|
||||
const chunkIndex = Math.floor(index / chunkSize);
|
||||
const chunk = chunks[chunkIndex] || [];
|
||||
chunks[chunkIndex] = chunk.concat([elem]);
|
||||
return chunks;
|
||||
}, [] as T[][]);
|
||||
369
src/ts/process/memory/supaMemory.ts
Normal file
369
src/ts/process/memory/supaMemory.ts
Normal file
@@ -0,0 +1,369 @@
|
||||
import { get } from "svelte/store";
|
||||
import type { OpenAIChat } from "..";
|
||||
import { DataBase, type Chat, type character, type groupChat } from "../../storage/database";
|
||||
import { tokenize, type ChatTokenizer } from "../../tokenizer";
|
||||
import { requestChatData } from "../request";
|
||||
import { cloneDeep } from "lodash";
|
||||
import { HypaProcesser } from "./hypamemory";
|
||||
import { stringlizeChat } from "../stringlize";
|
||||
|
||||
export async function supaMemory(
|
||||
chats:OpenAIChat[],
|
||||
currentTokens:number,
|
||||
maxContextTokens:number,
|
||||
room:Chat,
|
||||
char:character|groupChat,
|
||||
tokenizer:ChatTokenizer,
|
||||
arg:{asHyper?:boolean} = {}
|
||||
): Promise<{ currentTokens: number; chats: OpenAIChat[]; error?:string; memory?:string;lastId?:string}>{
|
||||
const db = get(DataBase)
|
||||
|
||||
currentTokens += 10
|
||||
|
||||
if(currentTokens > maxContextTokens){
|
||||
let coIndex = -1
|
||||
for(let i=0;i<chats.length;i++){
|
||||
if(chats[i].memo === 'NewChat'){
|
||||
coIndex = i
|
||||
break
|
||||
}
|
||||
}
|
||||
if(coIndex !== -1){
|
||||
for(let i=0;i<coIndex;i++){
|
||||
currentTokens -= await tokenizer.tokenizeChat(chats[0])
|
||||
chats.splice(0, 1)
|
||||
}
|
||||
}
|
||||
|
||||
let supaMemory = ''
|
||||
let hypaChunks:string[] = []
|
||||
let lastId = ''
|
||||
let HypaData:HypaData[] = []
|
||||
|
||||
if(room.supaMemoryData && room.supaMemoryData.length > 4){
|
||||
const splited = room.supaMemoryData.split('\n')
|
||||
let id = splited.splice(0,1)[0]
|
||||
const data = splited.join('\n')
|
||||
|
||||
if(arg.asHyper && (!id.startsWith("hypa:"))){
|
||||
supaMemory = ""
|
||||
|
||||
}
|
||||
else{
|
||||
if(id.startsWith("hypa:")){
|
||||
|
||||
if((!arg.asHyper)){
|
||||
return {
|
||||
currentTokens: currentTokens,
|
||||
chats: chats,
|
||||
error: "SupaMemory: Data saved in hypaMemory, loaded as SupaMemory."
|
||||
}
|
||||
}
|
||||
|
||||
HypaData = JSON.parse(data.substring(0,5).trim())
|
||||
if(!Array.isArray(HypaData)){
|
||||
return {
|
||||
currentTokens: currentTokens,
|
||||
chats: chats,
|
||||
error: "hypaMemory: hypaMemory isn't Array"
|
||||
}
|
||||
}
|
||||
|
||||
let indexSelected = -1
|
||||
for(let i=0;i<HypaData.length;i++){
|
||||
let i =0;
|
||||
let countTokens = currentTokens
|
||||
let countChats = cloneDeep(chats)
|
||||
while(true){
|
||||
if(countChats.length === 0){
|
||||
break
|
||||
}
|
||||
if(countChats[0].memo === HypaData[i].id){
|
||||
lastId = HypaData[i].id
|
||||
currentTokens = countTokens
|
||||
chats = countChats
|
||||
indexSelected = i
|
||||
break
|
||||
}
|
||||
countTokens -= await tokenizer.tokenizeChat(countChats[0])
|
||||
countChats.splice(0, 1)
|
||||
i += 1
|
||||
}
|
||||
if(indexSelected !== -1){
|
||||
break
|
||||
}
|
||||
}
|
||||
if(indexSelected === -1){
|
||||
return {
|
||||
currentTokens: currentTokens,
|
||||
chats: chats,
|
||||
error: "hypaMemory: chat ID not found"
|
||||
}
|
||||
}
|
||||
|
||||
supaMemory = HypaData[indexSelected].supa
|
||||
hypaChunks = HypaData[indexSelected].hypa
|
||||
|
||||
}
|
||||
else{
|
||||
let i =0;
|
||||
while(true){
|
||||
if(chats.length === 0){
|
||||
return {
|
||||
currentTokens: currentTokens,
|
||||
chats: chats,
|
||||
error: "SupaMemory: chat ID not found"
|
||||
}
|
||||
}
|
||||
if(chats[0].memo === id){
|
||||
lastId = id
|
||||
break
|
||||
}
|
||||
currentTokens -= await tokenizer.tokenizeChat(chats[0])
|
||||
chats.splice(0, 1)
|
||||
i += 1
|
||||
}
|
||||
|
||||
supaMemory = data
|
||||
currentTokens += await tokenize(supaMemory)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
if(currentTokens < maxContextTokens){
|
||||
chats.unshift({
|
||||
role: "system",
|
||||
content: supaMemory
|
||||
})
|
||||
return {
|
||||
currentTokens: currentTokens,
|
||||
chats: chats
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
async function summarize(stringlizedChat:string){
|
||||
|
||||
const supaPrompt = db.supaMemoryPrompt === '' ?
|
||||
"[Summarize the ongoing role story, It must also remove redundancy and unnecessary text and content from the output to reduce tokens for gpt3 and other sublanguage models]\n"
|
||||
: db.supaMemoryPrompt
|
||||
|
||||
let result = ''
|
||||
|
||||
if(db.supaMemoryType !== 'subModel'){
|
||||
const promptbody = stringlizedChat + '\n\n' + supaPrompt + "\n\nOutput:"
|
||||
|
||||
const da = await fetch("https://api.openai.com/v1/completions",{
|
||||
headers: {
|
||||
"Content-Type": "application/json",
|
||||
"Authorization": "Bearer " + db.supaMemoryKey
|
||||
},
|
||||
method: "POST",
|
||||
body: JSON.stringify({
|
||||
"model": db.supaMemoryType === 'curie' ? "text-curie-001" : "text-davinci-003",
|
||||
"prompt": promptbody,
|
||||
"max_tokens": 600,
|
||||
"temperature": 0
|
||||
})
|
||||
})
|
||||
|
||||
if(da.status < 200 || da.status >= 300){
|
||||
return {
|
||||
currentTokens: currentTokens,
|
||||
chats: chats,
|
||||
error: "SupaMemory: HTTP: " + await da.text()
|
||||
}
|
||||
}
|
||||
result = (await da.json()).choices[0].text.trim()
|
||||
}
|
||||
else {
|
||||
const promptbody:OpenAIChat[] = [
|
||||
{
|
||||
role: "user",
|
||||
content: stringlizedChat
|
||||
},
|
||||
{
|
||||
role: "system",
|
||||
content: supaPrompt
|
||||
}
|
||||
]
|
||||
const da = await requestChatData({
|
||||
formated: promptbody,
|
||||
bias: {}
|
||||
}, 'submodel')
|
||||
if(da.type === 'fail' || da.type === 'streaming' || da.type === 'multiline'){
|
||||
return {
|
||||
currentTokens: currentTokens,
|
||||
chats: chats,
|
||||
error: "SupaMemory: HTTP: " + da.result
|
||||
}
|
||||
}
|
||||
result = da.result
|
||||
}
|
||||
return result
|
||||
}
|
||||
|
||||
let hypaResult = ""
|
||||
|
||||
if(arg.asHyper){
|
||||
const hypa = new HypaProcesser()
|
||||
await hypa.addText(hypaChunks)
|
||||
const filteredChat = chats.filter((r) => r.role !== 'system' && r.role !== 'function')
|
||||
const s = await hypa.similaritySearch(stringlizeChat(filteredChat.slice(0, 4)))
|
||||
hypaResult = s.slice(0,4).join("\n\n")
|
||||
currentTokens += await tokenizer.tokenizeChat({
|
||||
role: "assistant",
|
||||
content: hypaResult
|
||||
})
|
||||
}
|
||||
|
||||
while(currentTokens > maxContextTokens){
|
||||
const beforeToken = currentTokens
|
||||
let maxChunkSize = maxContextTokens > 3500 ? 1200 : Math.floor(maxContextTokens / 3)
|
||||
let summarized = false
|
||||
let chunkSize = 0
|
||||
let stringlizedChat = ''
|
||||
let spiceLen = 0
|
||||
while(true){
|
||||
const cont = chats[spiceLen]
|
||||
if(!cont){
|
||||
currentTokens = beforeToken
|
||||
stringlizedChat = ''
|
||||
chunkSize = 0
|
||||
spiceLen = 0
|
||||
if(summarized){
|
||||
if(maxChunkSize < 500){
|
||||
return {
|
||||
currentTokens: currentTokens,
|
||||
chats: chats,
|
||||
error: "Not Enough Tokens"
|
||||
}
|
||||
}
|
||||
maxChunkSize = maxChunkSize * 0.7
|
||||
}
|
||||
else{
|
||||
const result = await summarize(supaMemory)
|
||||
if(typeof(result) !== 'string'){
|
||||
return result
|
||||
}
|
||||
|
||||
console.log(currentTokens)
|
||||
currentTokens -= await tokenize(supaMemory)
|
||||
currentTokens += await tokenize(result + '\n\n')
|
||||
console.log(currentTokens)
|
||||
|
||||
supaMemory = result + '\n\n'
|
||||
summarized = true
|
||||
if(currentTokens <= maxContextTokens){
|
||||
break
|
||||
}
|
||||
}
|
||||
continue
|
||||
}
|
||||
const tokens = await tokenizer.tokenizeChat(cont)
|
||||
if((chunkSize + tokens) > maxChunkSize){
|
||||
if(stringlizedChat === ''){
|
||||
|
||||
|
||||
if(cont.role !== 'function' && cont.role !== 'system'){
|
||||
stringlizedChat += `${cont.role === 'assistant' ? char.type === 'group' ? '' : char.name : db.username}: ${cont.content}\n\n`
|
||||
}
|
||||
}
|
||||
lastId = cont.memo
|
||||
break
|
||||
}
|
||||
stringlizedChat += `${cont.role === 'assistant' ? char.type === 'group' ? '' : char.name : db.username}: ${cont.content}\n\n`
|
||||
spiceLen += 1
|
||||
currentTokens -= tokens
|
||||
chunkSize += tokens
|
||||
}
|
||||
chats.splice(0, spiceLen)
|
||||
|
||||
if(stringlizedChat !== ''){
|
||||
const result = await summarize(stringlizedChat)
|
||||
|
||||
if(typeof(result) !== 'string'){
|
||||
return result
|
||||
}
|
||||
|
||||
const tokenz = await tokenize(result + '\n\n')
|
||||
currentTokens += tokenz
|
||||
supaMemory += result.replace(/\n+/g,'\n') + '\n\n'
|
||||
|
||||
let SupaMemoryList = supaMemory.split('\n\n')
|
||||
if(SupaMemoryList.length >= 5){
|
||||
const oldSupaMemory = supaMemory
|
||||
let modifies:string[] = []
|
||||
for(let i=0;i<3;i++){
|
||||
modifies.push(SupaMemoryList.shift())
|
||||
}
|
||||
hypaChunks.push(...modifies)
|
||||
|
||||
const result = await summarize(supaMemory)
|
||||
if(typeof(result) !== 'string'){
|
||||
return result
|
||||
}
|
||||
|
||||
modifies.unshift(result.replace(/\n+/g,'\n'))
|
||||
supaMemory = modifies.join('\n\n') + '\n\n'
|
||||
|
||||
currentTokens -= await tokenize(oldSupaMemory)
|
||||
currentTokens += await tokenize(supaMemory)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
chats.unshift({
|
||||
role: "system",
|
||||
content: supaMemory,
|
||||
name: "supaMemory"
|
||||
})
|
||||
|
||||
|
||||
|
||||
if(arg.asHyper){
|
||||
if(hypaResult !== ''){
|
||||
chats.unshift({
|
||||
role: "assistant",
|
||||
content: hypaResult
|
||||
})
|
||||
}
|
||||
|
||||
if(HypaData[0] && HypaData[0].id === lastId){
|
||||
HypaData[0].hypa = hypaChunks
|
||||
HypaData[0].supa = supaMemory
|
||||
}
|
||||
else{
|
||||
HypaData.push({
|
||||
id: lastId,
|
||||
hypa: hypaChunks,
|
||||
supa: supaMemory
|
||||
})
|
||||
}
|
||||
|
||||
return {
|
||||
currentTokens: currentTokens,
|
||||
chats: chats,
|
||||
memory: JSON.stringify(HypaData, null, 2),
|
||||
lastId: lastId
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
return {
|
||||
currentTokens: currentTokens,
|
||||
chats: chats,
|
||||
memory: lastId + '\n' + supaMemory,
|
||||
lastId: lastId
|
||||
}
|
||||
|
||||
}
|
||||
return {
|
||||
currentTokens: currentTokens,
|
||||
chats: chats
|
||||
}
|
||||
}
|
||||
|
||||
type HypaData = {id:string,supa:string,hypa:string[]}
|
||||
|
||||
Reference in New Issue
Block a user